fix: reject invalid benchmark uncertainty

Signed-off-by: Gregor Zeitlinger <gregor.zeitlinger@grafana.com>

Author: zeitlinger

.mise/tasks/generate_benchmark_summary.py

@@ -273,7 +273,7 @@ def score_interval(result: dict) -> tuple[float, float] | None:
         return None
     try:
         error = float(metric.get("scoreError"))
-        if not math.isnan(error):
+        if math.isfinite(error) and error >= 0:
             return score - error, score + error
     except (ValueError, TypeError):
         pass

.mise/tasks/test_generate-benchmark-summary.py

@@ -100,6 +100,12 @@ def test_zero_or_non_finite_scores_are_inconclusive(self):
             comparison_status(result(score=float("inf")), result()), "inconclusive"
         )
 
+    def test_invalid_fallback_uncertainty_is_inconclusive(self):
+        for error in (float("inf"), float("-inf"), -1.0):
+            head = result(error=error)
+            head["primaryMetric"].pop("scoreConfidence")
+            self.assertEqual(comparison_status(head, result()), "inconclusive", error)
+
     def test_missing_confidence_interval_is_inconclusive(self):
         head = result(score=106, error=None)
         self.assertEqual(comparison_status(head, result()), "inconclusive")